Future‐Proofing AI Models in IoMT Environment: Adversarial Dataset Generation and Defense Strategies

ABSTRACT The growth of the Internet of medical things (IoMT) poses significant threats to data integrity and patient safety by creating vulnerabilities that adversarial attacks can exploit. The sharing of sensitive data between medical equipment is greatly aided by IoT, also known as IoMT, in the medical industry. Considering the problem of adversarial attacks, this study introduces a novel dataset by simulating such attacks on an existing IoMT dataset, with a focus on robust adversarial dataset generation and advanced defence mechanisms. It proposes a comprehensive framework to future‐proof AI models operating in realistic IoMT environments consisting of a hybrid approach for adversarial attack detection and classification consists of two phases: the first phase trains three machine learning (ML) classifiers as an ensemble voting classifier and one deep learning (DL) classifier, deep neural network (DNN) and the second phase trains a hybrid ensemble mode based on the above two models. The dataset is preprocessed, and then a random sampling technique is applied to resample the data. The performance of the proposed framework is evaluated using metrics such as accuracy, recall, precision, F1‐score and a confusion matrix. The proposed hybrid framework achieved 94% accuracy in comparison with other ML, DL and ensemble models.